Article

Cite This: Ind. Eng. Chem. Res. 2018, 57, 16077−16083 pubs.acs.org/IECR

Continuous Cyclic Distillation for Binary Solvent Exchange: The

Batch Stack
Phillip C. Wankat*
School of Chemical Engineering Purdue University, West Lafayette, Indiana 47907-2100, United States

ABSTRACT: As the pharmaceutical industry adopts continuous

processes, batch distillation for solvent exchange will need to be
replaced with a continuous distillation process. Cyclic distillation is
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

adapted to solvent exchange, and a simplified calculation method

that does not assume constant molar overflow is developed for
binary systems. A three-stage cyclic distillation for exchange of
methanol with water designed for a bottoms product that is 90%
Downloaded via YEUNGNAM UNIV on July 3, 2023 at 04:17:03 (UTC).

water produces a 99.3% methanol distillate, uses 25.4% less solvent

than constant volume batch distillation, and is roughly equivalent to
a five-stage continuous distillation. When water is the feed solvent
and is exchanged with methanol, a three-stage cyclic distillation uses
54.4% and 40.0% less solvent than constant volume batch distillation
and a three-stage continuous distillation, respectively.

S olvent exchange or solvent swap involves replacing one

solvent with another while the solute of interest remains in
solution. This unit operation is frequently required in
pharmaceutical and fine chemical manufacturing when differ-
ent solvents are required for reaction or separation steps. The
separation processes commonly used for removing the original
solvent are batch distillation and diafiltration. Diafiltration,
essentially ultrafiltration with a membrane that does not pass
solute but has no selectivity for the solvents, is often employed
if the exchanging solvent is more volatile.1,2
When the feed solvent is more volatile than the exchange
solvent, batch distillation3,4 or preferably constant volume
batch distillation5 is commonly used for solvent exchange.
Constant volume batch distillation uses exchange solvent vapor
to evaporate solvents in the still. If constant molal overflow is
valid, then there is no change in the moles of solvent in the still
pot. If molar liquid densities are equal, the volume will be
constant. As will be shown later, the analysis can be done with
significantly less restrictive assumptions. Constant volume
batch distillation is now the standard in the pharmaceutical
industry since significantly less replacement solvent is
required.5−8 Placement of a column on top of the still pot to
increase the separation is fairly common.7
Solvent exchange by batch or continuous distillation is quite
different from traditional distillation. The purpose of solvent
exchange is to transfer a nonvolatile solute from one solvent to
another with no loss in solute while keeping the solute
dissolved throughout the process. Unlike batch solvent
exchange distillation, continuous large-scale solvent exchange
distillation does not appear to be widely used or studied. One
exception is a recent paper on continuous solvent exchange
distillation.9
Although mixed solvents are often used, it is easiest to
explain the continuous distillation process when there is a pure
feed solvent containing the solute and the replacement solvent
is also pure. As a specific example, consider methanol is the
feed solvent containing the solute and water is the replacement
solvent. The desired bottoms product is water plus a small
amount of methanol containing the solute and the desired
distillate is almost pure methanol containing no solute. In a
continuous system this is a very unusual distillation system.
The volatile portion of the feed (methanol) is the desired
distillate product and the water added to the column as steam
is the desired bottoms product. The distillation column can be
operated with no reflux and with direct steam addition without
a reboiler. The equipment looks like an absorber or a beer still
in a biorefinery and is described in detail elsewhere.9 If the feed
solvent is more volatile and there is no azeotrope, continuous
solvent exchange distillation with a sufficient number of stages
can produce essentially pure feed solvent as distillate and
essentially pure exchange solvent plus solute as the bottoms
product.9
Batch distillation is the original distillation process, and
simple batch stills have been traced back to the first century
AD.10 Continuous distillation is a much later development and
is often traced back to Aeneas Coffey in 1830.10 The
pharmaceutical industry has traditionally employed batch
equipment, but there is currently a push to develop continuous
processes.
There have been extensive developments to decrease energy
use and capital costs of continuous distillation. The
Special Issue: Richard Noble Festschrift
Received: May 31, 2018
Revised: July 30, 2018
Accepted: August 1, 2018
Published: August 16, 2018

© 2018 American Chemical Society 16077 DOI: 10.1021/acs.iecr.8b02449

Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research
intensification method of most interest for this study is the use
of cyclic distillation, which involves the use of vapor flow with
no liquid flow followed by liquid flow with no vapor flow and
without mixing the liquid on adjacent trays. Although each tray
operates with varying concentrations within a cycle, a cyclic
steady state is achieved in which each cycle repeats over and
over. Cyclic distillation was first developed by Cannon and his
co-workers11,12 who showed that cyclic distillation could
decrease the number of stages required to achieve a given
separation. The steady trickle of papers that followed Cannon’s
̂
papers is reviewed by Maleta et al.13 and Bildea etal.14 The use
of cyclic operation in distillation was delayed because
commercial systems were unable to achieve liquid transfer
without mixing the liquid on adjacent stages with a large
number of stages. In 2011 a successful system that uses a sluice
chamber and valves underneath each tray to hold liquid
temporarily until the trays have drained was announced.13−16
Results from a commercial beer still13 and pilot scale results17
are available.
Typically, the requirements of exchange distillation do not
require a large number of trays and a simpler arrangement
consisting of sparged tanks similar to constant volume batch
distillation systems can often be employed. This paper explores
the application of cyclic distillation to exchange distillation and
presents a simplified method for doing the calculations for
binary exchange using a spreadsheet.
■ PROCESS FOR CONTINUOUS CYCLIC
DISTILLATION EXCHANGE
The process for continuous cyclic distillation exchange can be
conceptualized as a series of constant volume batch tanks that
are connected in series. The steps in the continuous cyclic
distillation exchange process are illustrated in Figure 1 for a
Figure 1. Steps for a cyclic batch stack exchange system with three
stages.
16078
Article
three-stage system. The bottom tank (#1) is sparged with pure
replacement solvent vapor. The vapor from tank #1 is sent to
tank #2 and so forth. The vapor from the last tank (#3 in
Figure 1), the distillate product, consists of mainly the feed
solvent. Tank #1 receives its feed liquid from tank #2 at the
end of a cycle af ter the liquid in tank #1 has been totally
removed. The liquid from tank #1 is the final product, which
consists of the nonvolatile solute dissolved in close to pure
replacement solvent. At the end of the vapor flow period and
after tank #1 has been emptied, the liquid in tank #2 is fed to
tank #1. Additional tanks in the series have similar steps with
the last tank in the series (#3 in Figure 1) receiving the feed
(feed solvent plus solute) after its liquid has all been
transferred. Of course, we would like to do the liquid transfer
steps as quickly as possible. With a large number of stages the
process shown in Figure 1 will be cumbersome, but with only
two or three stages it is not much more difficult than operating
a constant volume batch distillation system. With only one
stage, this system is a constant volume batch distillation
system.
■ THEORY
The theoretical models for cyclic distillation were included in
the reviews mentioned previously.13,14 Exchange systems have
not been studied previously. Since the major process
simulators do not support the calculation of cyclic
distillation,14 interested users need to develop their own
code. The model closest to the model developed in this paper
is that of Toftegard and Jorgensen18 who integrated similar
differential equations with a fourth-order Runge−Kutta
method for constant molar overflow. All models reviewed13,14
assume constant molar holdup in the reboiler and on the trays
which is equivalent to assuming constant molar overflow.
The theory developed in this paper for continuous cyclic
distillation exchange is essentially a constant volume batch
distillation calculation on each stage, with the stages stacked on
top of each other; thus, the name batch stack. A key difference
with the usual batch distillation calculations is the inlet vapor
concentration varies throughout the cycle for stages above the
first stage. A second key difference is that molar units are
changed to latent heat units so that the assumption of constant
molar overflow is not necessary.
The molar vapor and liquid flow rates and the molar liquid
holdup are usually not constant since the latent heats per mole
are not equal. In their classic paper McCabe and Thiele19
realized that their assumption of constant molar overflow was
often only approximately valid. They included a short section
explaining in words how latent heat units could be employed
to make the flow rates almost completely constant. The
McCabe−Thiele latent heat unit method is explained in
detail20 with an example for the separation of methanol and
water, which is the example considered here for an exchange
system. For a binary system the latent heat fraction of the
component with the lower latent heat (e.g., methanol) is
defined as
xM(λM /λW )
*=
xM
xM(λM /λW ) + x W (1a)
λM and λW are the pure component latent heats at the
distillation operating pressure (1.0 atm) of methanol and
water, respectively. For water, the component with the higher
* = 1 − xM
latent heat, xW * or
DOI: 10.1021/acs.iecr.8b02449
Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research Article

xW Equations 5a and b are a consequence of using latent heat

* =
xW
xM(λM /λW ) + x W (1b) units, and eq 5c is a steady-state mass balance over an entire
cycle. In eq 5a,b,c and in the remaining equations I will follow
To convert from the latent heat fractions to mole fractions the normal custom for binary distillation that x, y, x*, and y*
*
xM without a subscript refer to the more volatile component.
xM = Substituting eq 5a and b and into eq 5c and solving for S*, we
* )(λM /λW ) − xM
(1 − xM *
obtain
xW* (λM /λW )
xW = 1 − xM = (xF* − xB*)F*
* *
x W (λM /λW ) + 1 − x W (1c,d) S* =
(yD* − yS*) (6)
The flow rates are converted as

ji λ zy ji * λM y
* zzzz
V * = V jjjyM M + yW zzz V = V */jjjyM
Equation 5c is valid in any set of units, but eqs 5a and b and

j λW z j λW z
6 will only be valid in molar units if constant molal overflow is
k { k {
+ yW valid. If enough stages are available, eq 6 provides an estimate
of the solvent flow rate required to obtain the desired bottoms
(2a,b) and distillate compositions.
Molar liquid holdups in the still pots P are converted by similar If sensible heat changes can be ignored, eq 5a requires that

ji λ zy ji * λM y
equations, the holdups P* in the still pots in Figure 1 must all be constant.

P* = P jjjxM M + x W zzz P = P*/jjjxM * zzzz

Thus, the Rayleigh equation is simplified to the equation for

j λW z j λW z
k { k {
+ xW constant volume batch distillation, but without requiring
constant molar overflow. For a batch still pot with pure
(2c,d) entering vapor S*, the resulting equation in latent heat units
is5,8,9
All flow rates and compositions including the vapor−liquid
equilibrium (VLE) data need to be converted to latent heat *
xinitial dxP*
units. The example20 illustrates converting VLE data to latent S* / P * = ∫x *
final y* (7a)
heat units. For the exchange system the nonvolatile solute was
assumed to not affect VLE. If the solute affects VLE, ternary P* is the constant holdup of material in the still pot. The vapor
equilibrium data will be required. of composition y* is in equilibrium with the liquid in the still
To write the external mass balances in latent heat units for pot xP*.
the batch stack exchange system at cyclic steady state, the flow If the entering vapor is not pure, it needs to be included in
rates need to be normalized with respect to the time of one the mass balance. The resulting equation is
cycle (flow step plus liquid transfer step). D*V , S*V , and V*V are *
xinitial dxP*
the continuous flow rates in latent heat units/time during S* / P * = ∫x *
vapor flow for distillate, entering solvent vapor, and vapor flow final y* − yin* (7b)
in the column, respectively. If tV is the time for vapor flow and
tL is the time for liquid transfer, the normalized flow rates D*, If the relative volatility is constant, α* = α, and y*in is
constant eqs 7a and 7b can be integrated analytically. The

ÄÅ ÉÑ
S*, and V* are
solution for eq 7a is well-known.5,8

1 ÅÅÅÅ xP*,initial ÑÑÑÑ

lnÅ Ñ+
D* = t VD V*/(t V + t L) S* = t VS V*/(t V + t L)

α* ÅÅÅÅÇ xP*,final ÑÑÑÑÖ

α* − 1 *
V * = t VV V*/(t V + t L) (3a,b,c)
S* / P * = (xP ,initial − xP*,final)
α*
For the liquid transfer the liquid latent heat unit average flow (8a)
rate is If the entering solvent vapor is not pure but is constant and
(latent heat units of liquid transferred) relative volatility is constant, eq 7b has a more complicated
L L* = analytical solution,

ij λ yz
tL

= (moles liquid transferred)jjjxM M + x W zzz/t L

1

j λW z
S* / P * =

k { l
[α* − yin*(α* − 1)]
o
o [α* − yin*(α* − 1)]xinitial
×m
(4a)
o
o
* − y*
o
n
in
and the normalized liquid flow rates are ln
* * − y*
|
[α* − yin (α* − 1)]xfinal
o
* )o
in

}
o
L* = t LL V*/(t V + t L) F * = t LFV*/(t V + t L)
o
o
~
* − xfinal
+ (α* − 1)(xinitial
B* = t LB V*/(t V + t L) (4b,c,d)
In molar units the transformations from the actual vapor flow (α* − 1)yin* * − y* /[α* − y*(α* − 1)]
xinitial in in
+ 2
ln
rate and the actual amount of liquid transferred will have the *
[α* − yin (α* − 1)] * − y* /[α* − y*(α* − 1)]
xfinal in in
same form as eqs 3a,b,c, 4a, and 4b,c,d. (8b)
The normalized flow rates in the cyclic exchange distillation
system satisfy the following equations, Equation 8b simplifies to eq 8a when yin* = 0. Equation 8b is
convenient for a system with a single constant volume batch
V * = D* = S* L * = B* = F * pot when the entering vapor is not pure. Unfortunately, this
* =
equation is less useful for a cascade of batch pots since yk,in
F *xF* + S*yS* = D*xD* + B*xB* (5a,b,c) *
yk−1,out is not constant for stages with k > 1. Numerical
16079 DOI: 10.1021/acs.iecr.8b02449
Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research Article

Table 1. Results for Example 1a

steps * final
xM * final
yM * initial
xM * initial
yM * average
xM * average
yM V*/L*
Run 1, Stage 1
5 (last) 0.0879 0.3872 0.1784 0.5445 0.1332 0.4813 0.192
step 4 0.1784 0.5445 0.2941 0.6512 0.2363 0.6078 0.192
step 3 0.2941 0.6512 0.4290 0.7395 0.3616 0.7071 0.192
step 2 0.4290 0.7395 0.5774 0.8091 0.5032 0.7733 0.192
step 1 0.5892 0.8091 0.7402 0.8867 0.6588 0.8488 0.192
liquid flow total: 0.960
Run 1, Stage 2
step 5 0.7402 0.8867 0.8221 0.9240 0.7811 0.9053 0.192
step 4 0.8221 0.9240 0.8859 0.9532 0.8540 0.9386 0.192
step 3 0.8859 0.9532 0.9355 0.9736 0.9107 0.9628 0.192
step 2 0.9355 0.9736 0.9752 0.9895 0.9554 0.9812 0.192
step 1 0.9752 0.9895 1.0028 1.0 0.9890 0.9954 0.192
total = V*/L*: 0.960
Run 2, Stage 1
step 5 0.0879 0.3872 0.1781 0.5442 0.1331 0.4811 0.1916
step 1 0.5761 0.8084 0.7384 0.8859 0.6572 0.8480 0.1916
liquid flow
Run 2, Stage 2
step 5 0.7384 0.8859 0.8199 0.9230 0.7791 0.9044 0.1916
step 1 0.9724 0.9884 0.99998 0.99999 0.9862 0.9942 0.1916
total V*/L* for run 2: 0.958
a
Apparatus is shown in Figure 1 and is operated in a cyclic fashion. Feed is 100 kmol/h with methanol mole fraction = 1.0 on a solute free basis,
entering steam is pure saturated vapor, and bottoms is 0.10 mole fraction methanol, p = 1.0 atm.

integration of eqs 7a and 7b in a batch stack is illustrated with a first example, a value of V*/L* = 0.96 (1.053 times larger than
detailed example in the next section. the minimum) is used for the first trial.

■
Equilibrium data were obtained for the NRTL correlation
EXCHANGE WITH FEED SOLVENT MORE from Aspen Plus. The y−x pair values were transferred to a
VOLATILE spreadsheet and transformed to y*−x* pairs with eqs 1a and
1b. Linear interpolation was used to find y* values for any x*
The equations developed previously are valid regardless of
value. For the batch stack the bottom stage, k = 1, in the
which solvent is more volatile. However, batch distillation for
column has pure replacement solvent vapor (steam in this
exchange is most commonly employed when the feed solvent is
example, yS = 0) entering throughout the vapor flow step; thus,
the more volatile component. A continuous exchange system
eq 7a applies to the bottom stage. Since the final condition for
can exchange the more volatile feed solvent using significantly
stage 1 is known (x*B = 0.0879 in this example), eq 7a can be
less replacement solvent than batch distillation as long as there
integrated backward from step Nstep to step 1 to find the
are two or more stages in the continuous system.9 In this
*
unknown value of x1,1,initial using Simpson’s rule. The value of
section the application of cyclic operation with a batch stack
x*1,initial is determined by forcing the calculated value of V*/L*
for exchange of a more volatile feed solvent is analyzed with
= 0.96/Nstep for each step.
detailed examples.
At the end of vapor flow, represented as Nstep steps, the
Example 1: Batch stack for replacement of pure methanol
liquid in stage 1 becomes the bottom product. The liquid in
feed solvent with pure water replacement solvent. The solute is
stage 2 is transferred into stage 1 without mixing with the
nonvolatile and is assumed to not affect the vapor−liquid
liquid originally in stage 1. Thus, the final liquid concentration
equilibrium of methanol and water. Feed rate F = 100 kmol/h
in stage 2 (at step Nstep) is equal to the initial liquid
(solute-free basis), feed mole fraction (solute-free basis) xF = 1,
*
concentration in stage 1 (at step 1): x2,Nstep,final *
= x1,1,initial. Since
replacement solvent mole fraction yS = 0, p = 1.0 atm, desired
the vapor entering stage 2 comes from stage 1, it varies in
bottoms xB = 0.1, and desired distillate yD ≈ 1. Find the
composition and eq 7b written for a cascade is needed
required replacement solvent flow rate and the concentration
of the distillate for a two-stage cyclic solvent exchange xk*,initial dx L*
distillation system.
For this system λM/λW = 0.8675, which gives, F* =
(V */L*)k = ∫x * yk* − yk*− 1
k ,final (9)
F( λ
xM, F λM
W
)
+ x W, F = F(λM/λW) = 86.75, xF* = xF = 1, yS* = yS
with k = 2 for stage 2. To integrate eq 9 numerically, the values
= 0, x*B = 0.0879, and y*D ≈ 1. Then from eq 6, for xB = 0.1, S* of y*k−1 need to be known during the vapor flow step. These
= (xF* − xB*)F *
= (1 − 0.0879)(86.75)
= 79.13, and since yS,M = 0, S = values were calculated at specified values of V*/L* from the
(y* − y*)
D S
(1 − 0) integration of stage 1. Thus, the same values of V*/L* must be
S* = 79.13. In latent heat units for a perfect separation with a used for every stage of the cascade. With a spreadsheet it is
large number of stages: L* = F* = 86.75, V* = S* = 79.13, L*/ convenient to use goal seek to force identical values of V*/L*.
V* = 86.75/79.13 = 1.096, and V*/L* = 0.912. If only a few For this example the vapor flow step was divided into five
stages will be used, a larger value of V*/L* is needed. In the segments, Nstep = 5, with (V*/L*)step = 0.96/5 = 0.192. For the
16080 DOI: 10.1021/acs.iecr.8b02449
Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research Article

Table 2. Comparison of Batch Stack Results for Exchange of Pure Methanol Feed with Water with Continuous Flow System
Simulationsa
SS continuous flow column (Aspen) 2− SS continuous flow column (latent heat
cyclic continuous (batch stack) latent heat calculation 10 stages9 calc)
Vfeed/F = S/F yD Vfeed/F yD Vfeed/F yD
1 stage 1.072 (5 segments) 0.765 2.0375 0.421 2.176 0.419
1.070 (10 segments) 0.766
2 stages 0.831 0.958 0.968 0.824 1.124 0.812
3 stages 0.798 0.993 0.824 0.943 0.974 0.937
5 stages 0.781 0.994
a
Vfeed/F = S/F = steam/feed. Same feed and bottoms as in Table 1.

other stages the beginning, ending, and average y values over

each time segment from the stage below were used in ( λ
Check: Calculate B* from eq 2a: B*calc = B xM, B λM + x W, B =
W
)
Simpson’s rule, which takes the following form for stage k 87.77[0.1(0.8675) + 0.9)] = 86.61.

ÄÅ
during step j, Comparison with the liquid flow rate determined from

ij V * yz (xk*, j ,initial − xk*, j ,final ÅÅÅÅ

constant latent heats B* = F* = 86.75 shows there is numerical
jj zz = ÅÅ
Å
Å
k L* { k , j ÅÅ (yk , j ,initial − yk*− 1, j ,initial )
1 error of 0.14 kmol/h, which is 0.16%. As a result, the external

ÅÇ
mass balance does not close: F + Vin − D − B = 100 + 83.11 −
ÉÑ
6 *

ÑÑ
95.11 − 87.77 = 0.23, although the discrepancy is quite small.
ÑÑ
ÑÑ
If the solvent flow rate is increased, a two-stage system can

(yk , j ,final − yk − 1, j ,final ) ÑÑÑÑ

4 1
+ +
* *
(yk , j ,av − yk − 1, j ,av )
If k = 1, all yk−1 = yS (=0 in this example). The integration steps
are done backward from j = Nstep (5 in this example) to j = 1.
When the j = 1 step is finished and xk,1,initial * has been
determined, the liquid composition for the stage above is
x*k+1,Nstep,final = x*k,1,initial. Integrating from the feed stage and
stepping downward is significantly more difficult since the
vapor mole fractions throughout the cycle are not known.
Results are shown in Table 1 for all five time steps for stages
1 and 2. Note that since the integration is done backward, the
calculation is done first for vapor flow step 5, and then
sequentially for the earlier flow steps.
Obviously, the feed for run 1 (x*M,initial in stage 2, step 1)
cannot be greater than 1.0, and trial-and-error is required to
find the V*/L* that converges. The requirement for
convergence is
produce a bottoms product with less water. With a steam to
ÑÖ
* * feed ratio of Sinit/F = 0.9465 (S*/W* = 1.09) the bottoms (on
a solute-free basis) is 0.01 mole fraction methanol (x* =
(10)
0.00869) for a feed that is 0.9925 mole fraction methanol. The
distillate is 0.9974 mole fraction methanol, which is higher
than the feed concentration. With a slight increase in the
steam/feed ratio, a pure methanol feed would be exchanged
with water and the bottoms would be 0.01 mole fraction
methanol. The ratio of steam/feed increased from 0.831 when
the bottoms product was 0.10 mole fraction methanol to
0.9465 for a 0.01 mole fraction methanol product, which is a
13.9% increase.
Example 2: The calculation in Example 1 was repeated for a
three-stage system. With V*/L* = 0.920 the calculated initial
feed latent heat fraction was x* = 0.9951 (ε* = 0.0049). The
flow rates and mole fractions are L* = F* = 86.75 (time
averaged over the vapor flow period), V* = S* = 79.81, Sinit =
79.81 kmol/h, S/F = steam/Feed = 0.798, B = L1 = 87.92, xB =
0.10, D = V3 = 91.89, yD,avg = 0.9925, x3,1,initial = 0.9958 (ε =
0.0042).

■
xk*,1,initial = xF* ± ε*, and xk ,1,initial = xF ± ε (11)

where ε is a suitable small number (e.g., ε = 0.0001) and k = 2 COMPARISON WITH NORMAL EXCHANGE
for this example. Run 2 in Table 1 shows the converged result DISTILLATION AND CONSTANT VOLUME BATCH
(ε* = 0.00002) for V*/L* = 0.958. Since L* = F* = 86.75 DISTILLATION
where L* and F* are the normalized flow rates, the normalized Comparisons of the batch stack system with other methods of
vapor velocity V* = 83.11. The entering vapor is pure water, doing the same exchange are instructive. With one stage the
the compound with the higher latent heat. Then from eq 2a,b S batch stack exchange system becomes equivalent to constant
= Vin = V* = 83.11 kmol/h, and the ratio of steam/feed = volume batch distillation with repeated charges of feed. Results
0.831. for the solution of eq 7 using Simpson’s rule with 10
A methanol mass balance at cyclic steady state can be used subdivisions for a feed of 100 kmol/h of pure methanol gives
to find the methanol in the distillate, the following flow rates: Steam = 107.02 (steam/feed ratio =
1.07), B = 87.92, and D = 119.1 kmol/h. The values for S/F
yD* = (F *xF* − L*xB*)/V * (12) and yD with 5 and 10 segments are compared in Table 2. The
difference in yD is 0.13% and the difference is S/F is 0.19%. For
The results are y*D = 0.952 and yD = 0.958. The distillate flow
exchange of methanol with water, five segments in the
rate D* = V* = 83.11, and from eq 2b D = D*/
numerical integration with Simpson’s rule appear to be
( λ
y* M + 1 − y* = 95.11 kmol/h.
D λW D ) sufficient.
Aspen Plus was used to simulate a continuous one stage
Since B* = L* = 86.75, the value of B is B = B*/
system as a flash unit with a vapor feed of pure steam and a
( λ
)
xB* M + 1 − xB* = 87.77 kmol/h.
λW
liquid feed of saturated liquid methanol. Results9 for a
16081 DOI: 10.1021/acs.iecr.8b02449
Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research
Table 3. Comparison of Results for Exchanging Pure Water Feed with Replacement Solvent Pure Methanola
SS continuous flow column (Aspen) (2
cyclic continuous (latent heat calc)
Vfeed/F yD,W
1 stage 5.45 (5 segments) 0.167
5.53 (10 segments) 0.164
2 stages 3.23 0.287
3 stages 2.52 0.352
a
Final solution containing the solute is 90% methanol and 10% water (solute-free basis). p = 1.0 atm.
continuous flow exchange distillation system with N ≥ 2 are
also listed in Table 2.
The latent heat calculations for a steady-state exchange
column are also shown in Table 2 to allow comparison of the
accuracy of these calculations compared to the Aspen Plus
simulations. Since the latent heat calculation used the results of
the Aspen Plus NRTL VLE correlation, differences in the two
calculations are due to numerical errors and error in the
assumption that constant latent heat unit flows and holdups
are constant. Since sensible heat effects are not included in the
latent heat calculation, we would expect some error in the
results. The values of yD are reasonably accurate, but the latent
heat calculation values for Vfeed/F are systematically high. If
this trend is also true for the batch stack system, the
predictions of Vfeed/F in Table 2 for the batch stack are high,
which is conservative.
■ EXCHANGE WITH FEED SOLVENT LESS VOLATILE
THAN REPLACEMENT SOLVENT
Solvent exchange will often be required when the feed solvent
is less volatile. Although batch distillation has not usually been
employed for this exchange, a continuous exchange column
could often do the exchange using less solvent than the
alternative separation method continuous constant volume
diafiltration.9 The calculation method for a batch stack system
when the feed is the less volatile component is essentially the
same as in Examples 1 and 2 except Simpson’s method was
written in terms of the less volatile component. Example 3
illustrates exchange results when water is the feed solvent and
methanol is the replacement solvent.
Example 3. The feed is pure water on a solute-free basis at a
flow rate of 100 kmol/h. The solvent is pure methanol vapor. A
bottoms that contain xw = 0.1 mol fraction water (x*w =
0.1135) is desired. The equilibrium data at 1.0 atm was
identical to that used in Examples 1 and 2 and was converted
to latent heat units. Different values of replacement solvent
flow rate were tried until one, two, and three stages could do
the separation with a small error. The results for comparison
are reported in Table 3.
A comparison with steady-state continuous flow distillation
is shown in Table 3 for exchanging a water feed with
replacement solvent methanol. As expected, significantly more
solvent is required for exchange when the less volatile solvent
is the feed solvent than when the more volatile solvent is the
feed solvent. The results in Tables 2 and 3 show that single
stage continuous flow systems (e.g., sparged stirred tank
distillations) are remarkably inefficient particularly when the
replacement solvent is less volatile. To check for numerical
error, the one-stage cyclic system was repeated with 10 steps in
Simpson’s rule. The resulting values (shown in Table 3) for
Vfeed/F and yD differ from the results with 5 steps by 1.5% and
1.8%, respectively.
16082
Article
SS continuous flow column (latent heat
and 3 stages9) calc)
Vfeed/F yD,W Vfeed/F yD,W
23.15 0.0402 21.24 0.0417
6.45 0.140 6.05 0.147
4.20 0.218 3.95 0.224
■ DISCUSSION
Because each stage is designed to hold liquid without weeping,
start-up can be accomplished quickly. The top stage should
receive the feed. For the other stages mix the feed and
replacement solvent in the correct amounts to make the
concentrations calculated for each stage at the beginning of
each cycle. These concentrations are calculated for each stage
above the bottom from the value of x* initial for step 1 (for
example, in Table 1) using eqs 1c and 1d. Since start-up can be
very rapid, the batch stack method is well suited to campaigns.
To optimize the capacity, it is desirable to make the ratio tV/
tL as large as possible by transferring the liquid as quickly as
possible. With few stages, sequential transfer, shown in Figure
1, does not make tL too large, but this method is problematic
with a large number of stages. If a large number of stages is
required, other equipment should be considered.13−17 Because
the entire cross-sectional area of each stage is available for
vapor flow, higher vapor velocities can be employed than in
standard distillation equipment.
The analysis assumed that all liquid was drained from each
stage before the liquid from the stage above was added. This
assumption can be relaxed by including a mass balance to mix
the retained liquid with the added liquid on each stage. In this
paper vapor flow was stopped during the period of liquid
transfer to allow for a relatively simple, but accurate, analysis of
the separation. Continuing vapor flow during the liquid
transfer steps will simplify the operation if it does not interfere
with liquid transfer, but continuous vapor flow will complicate
the analysis significantly during the liquid flow period.
Results in Table 2 show that for exchanging methanol as the
feed solvent with water as the replacement solvent a three-
stage batch stack system can reduce the amount of solvent
compared to a constant volume batch system by 25.4% and
produce a distillate product that is 99.3% methanol, which may
not require any additional purification to be reused in the
process. For the more difficult exchange of water as the feed
solvent with methanol as the replacement solvent, Table 3
shows that a three-stage batch stack system can reduce the
amount of solvent compared to a constant volume batch
system by 54.4% and more than double the distillate
concentration.
Currently, the most common alternative for batch
distillation when the replacement solvent is more volatile is
to use batch constant volume diafiltration (Foley1 and
Schwartz2). To exchange a solvent and obtain a product that
contains 10% of the original solvent in a batch diafiltration
system requires a replacement solvent/feed ratio of 2.303, and
to obtain 1% of the original solvent requires a diafiltration ratio
of solvent/feed = 4.605 (e.g., Schwartz2). Since methanol is
quite volatile compared to water, these ratios are less than
required for a batch stack with three stages. The diafiltration
results are independent of the relative volatility of the feed and
DOI: 10.1021/acs.iecr.8b02449
Ind. Eng. Chem. Res. 2018, 57, 16077−16083
Industrial & Engineering Chemistry Research Article

replacement solvents. If the relative volatility of the feed
solvent compared to the replacement solvent is less than 1 but
closer to 1 than the methanol−water system, a batch stack
system may well require less replacement solvent than batch
constant volume diafiltration. For example, for pure water as
the original solvent exchanged for ethanol as the replacement
solvent with two stages and a bottoms product that is 0.10
mole fraction water (solute-free basis) S*/W* = 1.931, Sin/F =
2.035, and yD,W = 0.451. In this case the batch stack system is
likely to be competitive with batch constant volume
diafiltration. Continuous constant volume diafiltration with
one stage is not very effective since the two outlet streams must
have equal mole fractions. As a result, to reduce the
concentration of pure feed solvent to 10% Sin/F = 9.0.
Another difficulty with diafiltration is retention of solutes is
often less than 1.0 for relatively low molecular weight solutes.
The model presented here could easily be used for other
binary applications of cyclic distillation. Liquid feed can be
input at any time during the cycle although it may be most
convenient to add the feed during the liquid transfer step. The
optimum point to add feed will probably be when the feed
matches the liquid concentration in the column. Reflux is
added at the liquid transfer step. Since the concentration of the
vapor leaving the top stage varies during the cycle, the reflux
could be the same or different from the distillate product.
(6) Li, Y.-E.; Yang, Y.; Kalthod, V.; Tyler, S. M. Optimization of
solvent chasing in API manufacturing process: Constant Volume
Distillation. Org. Process Res. Dev. 2009, 13 (1), 73−77.
(7) Papadakis, E.; Tula, A. K.; Gani, R. Solvent selection
methodology for pharmaceutical processes: Solvent swap. Chem.
Engr. Rsch & Des 2016, 115, 443−461.
(8) Wankat, P. C Separation Process Engineering, 4th ed.; Prentice-
Hall: Upper Saddle River, NJ, 2017; Chapter 9.
(9) Fazlollahi, F.; Wankat, P. C. Novel Solvent Exchange Distillation
Column. Chem. Eng. Sci. 2018, 184, 216−228.
(10) Davies, J. T. Chemical Engineering: How Did it Begin and
Develop? In History of Chemical Engineering; Furter, W. F., Ed.;
Advances in Chemistry Series; American Chemical Society:
Washington, D.C., 1980, 190, 15−43.
(11) Gaska, R.; Cannon, M. R. Controlled cycling distillation in
sieve and screen plate towers. Ind. Eng. Chem. 1961, 53, 630−631.
(12) McWhirter, J. C.; Cannon, M. R. Controlled cycling distillation.
Ind. Eng. Chem. 1961, 53, 632−634.
(13) Maleta, V. N.; Kiss, A. A.; Taran, V. M.; Maleta, B. V.
Understanding Process Intensification in Cyclic Distillation Systems.
Chem. Eng. Process. 2011, 50 (7), 655−664.
(14) Bîldea, C. S.; Patrut, C.; Jorgensen, S. B.; Abildskov, J.; Kiss, A.
A. Cyclic distillation technology − a mini-review. J. Chem. Technol.
Biotechnol. 2016, 91 (5), 1215−1223.
(15) Maleta, B.; Maleta, O. Mass exchange contact device. US Patent
8,158,073, 2012.
(16) Kiss, A. A.; Bîldea, C. S. Revive Your Columns with Cyclic

■
SUMMARY
Solvent exchange in the pharmaceutical industry has typically
used batch distillation, most recently constant volume batch
distillation. The separation has typically been done in sparged
tanks. Converting batch solvent exchange to continuous
operation with the batch stack system will use very similar
equipment with few stages. The resulting cyclic steady-state
system will be easy to start-up, require less solvent, and
produce a more concentrated distillate product than constant-
volume batch distillation or continuous exchange distillation
with the same number of stages. When the feed solvent is more
volatile than the replacement solvent, the distillate product
may be close to pure feed solvent that does not need further
separation to be reused in the process.
Distillation. Chem. Engr. Progress 2015, 111 (12), 21−27.
(17) Maleta, B. V.; Shevchenko, A.; Bedryk, O.; Kiss, A. A. Pilot-
Scale Studies of Process Intensification by Cyclic Distillation. AIChE J.
2015, 61 (8), 2581−2591.
(18) Toftegaard, B.; Jorgensen, S. B. Design algorithm for periodic
cycled binary distillation columns. Ind. Eng. Chem. Res. 1987, 26 (5),
1041−1043.
(19) McCabe, W. L.; Thiele, E. W. Graphical rectifying column
calculations. Ind. Eng. Chem. 1925, 17 (6), 605−611.
(20) Wankat, P. C. Equilibrium Staged Separations; Prentice Hall;
Upper Saddle River, NJ, 1988; pp 189−197.

■ AUTHOR INFORMATION
Corresponding Author
*Phone: 765-494-4059. E-mail: wankat@ecn.purdue.edu.
ORCID
Phillip C. Wankat: 0000-0003-0352-0397
Notes
The author declares no competing financial interest.

■ REFERENCES
(1) Foley, G. Membrane Filtration. A Problem Solving Approach with
MATLAB; Cambridge University Press: Cambridge, U.K., 2013.
(2) Schwartz, L. Diafiltration for desalting or buffer exchange.
Bioprocess International 2003, No. 5, 43−49.
(3) Diwekar, U. W. Batch Distillation: Simulation, Optimal Design and
Control, 2nd ed.; Taylor and Francis: Washington, DC, 2012.
(4) Sorensen, E. Design and Operation of Batch Distillation. In
Distillation: Fundamentals and Principles; Gorak, A.; Sorensen, E., Eds.;
Elsevier: London, U.K., 2014; Chapter 5.
(5) Gentilcore, M. J. Reduce solvent usage in batch distillation.
Chem. Engr. Progress 2002, 98 (1), 56−59.

16083 DOI: 10.1021/acs.iecr.8b02449

Ind. Eng. Chem. Res. 2018, 57, 16077−16083